一种改进的基于SVM的SAR目标及阴影图像分割方法

SVM-based improved segmentation algorithm of SAR targets and shadow images

提出了一种基于支持向量机(support vector machine,SVM)的合成孔径雷达(synthetic aperture radar,SAR)目标及阴影图像的改进分割方法。利用分类的思想对SAR图像进行分割,其中分类器是通过循环不断更新训练样本的方式完成训练,循环次数由计算相邻两次分割图像熵的差值来控制。用DARPA(defense advanced research project agency)和Sandia实验室提供的实测数据进行分割实验。结果表明,所提算法得到的分类器性能更加优越,同时能够减少初始分割中阈值的选取对分类器性能的影响,有效地提高了SAR目标及阴影图像的分割质量。

An improved algorithm,which is based on support vector machine,is proposed for synthetic aperture radar（SAR） target and shadow image segmentation.A classification idea is used to perform SAR image segmentation.Training samples sent to support vector machine（SVM） are updated continuously by iterative processing.These iterations are repeated until the convergence,which is determined by checking the relative change of the entropy between two consecutive segmented images.The algorithm is applied to SAR imagery coming from defense advanced research project agency（DARPA） and Sandia Laboratory.Experimental results show that the classifier performance acquired from this algorithm is much better.Besides,it also dramatically reduces the influence of the choice of initial segmentation thresholds on the classifier performance and greatly increases the segmentation quality of SAR target and shadow image.